Eliminate wrong assumptions about "terminal" nodes

The process graph building logic in Python client quite heavily works from assumption about the equivalence

"end node" = "result node" = "save_result"

which works for most use cases, but that's not fully in line with full openEO API.

also see https://github.com/Open-EO/openeo-geopyspark-driver/issues/424#issuecomment-2199780221:

Some more insights from https://github.com/Open-EO/openeo-processes/issues/279 and https://github.com/Open-EO/openeo-api/issues/427 about different kinds of terminal nodes in the openEO API

• a process graph can have multiple "end nodes": leaf nodes that are not input to another node
• a process graph must have one "result node": node with "result": True.
  • The result node is actually just important in the the context of a "callback" (in a "parent" process like apply, reduce_dimension, ...) to determine the value that has to be used by the parent process
  • The "result" property has little meaning for "top-level" graphs
• the result node does not have to be an "end node". Result nodes can be input to subsequent nodes
• a "save_result" node neither has to be "end node" or "result node" to be executed

Start by locating the process graph building logic in the Python client and compare its terminal-node handling with the openEO API distinctions described in the issue and linked discussions. Done means end nodes, result nodes, and save_result nodes are handled independently, including graphs with multiple end nodes and result nodes that feed later nodes.